[ZenPsycho]

if you can infer a schema, I'd almost prefer a column oriented arrangement.

   "users": 
        {
         "first":["Homer","Hank","Peter"], 
         "last":["Simpson","Hill","Griffin"]
         },

then the information about the original structure can be restored by a set of object paths which need to be "rotated" from column to row orientation. ["users"]

saving a few characters in the process. though I see that the advantage of this system is supposedly that it can handle any sort of shape of data, not just ones with a fixed schema. I've been trying to figure out how trang [http://www.thaiopensource.com/relaxng/trang.html] does its schema inference trick. (it turns a set of xml files into a relaxNG schema). If you have a schema for a JSON file, it's knowledge you can apply to algorithmically creating really efficient transformations.

[dogada]

Exactly, ZenPsycho, the purpose of RJSON is to convert data with dynamic schema. Fields with default values are omitted often, for example if most of data have 'private' property set to False, it have sense to output it only for 1% of objects with 'private' set to True. This issue is addressed in RJSON.